On Monday, June 9, 2014 2:57:26 PM UTC+5:30, Steven D'Aprano wrote:

>    http://en.wikipedia.org/wiki/Landauer's_principle

Hey thanks for that!
Always thought something like this should exist but did not know what/where/how!

> On Sun, 08 Jun 2014 23:32:33 -0700, Rustom Mody wrote:

> > On Monday, June 9, 2014 9:50:38 AM UTC+5:30, Steven D'Aprano wrote:
> >> On Sun, 08 Jun 2014 19:24:52 -0700, Rustom Mody wrote:
> >> > On Monday, June 9, 2014 7:14:24 AM UTC+5:30, Steven D'Aprano wrote:
> >> >> CPU technology is the triumph of brute force over finesse.
> >> > If you are arguing that computers should not use millions/billions of
> >> > transistors, I wont argue, since I dont know the technology.
> >> No. I'm arguing that they shouldn't convert 90% of their energy input
> >> into heat.
> > Strange statement.
> > What should they convert it into then?

> Useful work, duh.

> Everything *eventually* gets converted to heat, but not immediately. 
> There's a big difference between a car that gets 100 miles to the gallon, 
> and one that gets 1 mile to the gallon. Likewise CPUs should get more 
> "processing units" (however you measure them) per watt of electricity 
> consumed.

> See, for example:

> http://www.tomshardware.com/reviews/fx-power-consumption-efficiency,3060.html

> http://en.wikipedia.org/wiki/Performance_per_watt

> Quote:

>     Theoretically, room-temperature computer memory operating 
>     at the Landauer limit could be changed at a rate of one 
>     billion bits per second with only 2.85 trillionths of a 
>     watt of power being expended in the memory media. Modern 
>     computers use millions of times as much energy.

Right so we are still very much in theoretical zone.
As the next para there says:

| If no information is erased, computation may in principle be achieved
| which is thermodynamically reversible, and require no release of
| heat. This has led to considerable interest in the study of reversible
| computing.

Particularly interesting as no-information-erasure corresponds to functional
(or maybe relational) programming. Of course still all theoretical.

> Much to my surprise, Wikipedia says that efficiency gains have actually 
> been *faster* than Moore's Law. This surprises me, but it makes sense: if 
> a CPU uses ten times more power to perform one hundred times more 
> computations, it has become much more efficient but still needs a much 
> bigger heat sink.

That was essentially my point


Reply via email to